A Study on Intrusion Detection Model Based on Hybrid Classifier

摘要

In order to improve the accuracy of classification problem in intrusion detection, a hybrid classifier which was composed by KPCA, BPNN and QGA, has been proposed in this paper. In the hybrid classifier, KPCA was used to reduce dimensions, and then QGA was used to search the best parameters for BPNN. BPNN which has been got the best weights matrix and thresholds by QGA, was used to train classification model. The main core factors of original dataset can be preserved by KPCA, and greatly reduced the computations. The weakness of BPNN, which was usually easy to get stuck in local minimum, can be solved by QGA. Finally, the effectiveness of hybrid classifier was proved by experiments. Compared with traditional methods, the hybrid classifier has better performance in reducing the classify errors.